oota-llvm.git
9 years agoRemove getSubtargetImpl calls from FastISel, we can get it from
Eric Christopher [Wed, 8 Oct 2014 23:38:33 +0000 (23:38 +0000)]
Remove getSubtargetImpl calls from FastISel, we can get it from
the MachineFunction where it's already cached.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219366 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix typo in comment
Robin Morisset [Wed, 8 Oct 2014 23:30:45 +0000 (23:30 +0000)]
Fix typo in comment

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219365 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSink DwarfUnit::addSectionDelta into DwarfCompileUnit, the only place it's needed.
David Blaikie [Wed, 8 Oct 2014 23:30:05 +0000 (23:30 +0000)]
Sink DwarfUnit::addSectionDelta into DwarfCompileUnit, the only place it's needed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219364 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Rename AVX512_masking* to AVX512_maskable*
Adam Nemet [Wed, 8 Oct 2014 23:25:39 +0000 (23:25 +0000)]
[AVX512] Rename AVX512_masking* to AVX512_maskable*

No functional change.

This is the current AVX512_maskable multiclass hierarchy:

                 maskable_custom
                    /       \
                   /         \
          maskable_common   maskable_in_asm
            /         \
           /           \
      maskable        maskable_3src

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219363 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Intrinsics for vextract*x4
Adam Nemet [Wed, 8 Oct 2014 23:25:37 +0000 (23:25 +0000)]
[AVX512] Intrinsics for vextract*x4

This adds the Pat<>'s for the intrinsics.  These are necessary because we
don't lower these intrinsics to SDNodes but match them directly.  See the
rational in the previous commit.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219362 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Add asm-only support for vextract*x4 masking variants
Adam Nemet [Wed, 8 Oct 2014 23:25:33 +0000 (23:25 +0000)]
[AVX512] Add asm-only support for vextract*x4 masking variants

These derive from the new asm-only masking definitions.

Unfortunately I wasn't able to find a ISel pattern that we could legally
generate for the masking variants.  The problem is that since the destination
is v4* we would need VK4 register classes and v4i1 value types to express the
masking.  These are however not legal types/classes in AVX512f but only in VL,
so things get complicated pretty quickly.  We can revisit this question later
if we have a more pressing need to express something like this.

So the ISel patterns are empty for the masking instructions and the next patch
will add Pat<>s instead to match the intrinsics calls with instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219361 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Move DAG for all-zero node to X86VectorVTInfo
Adam Nemet [Wed, 8 Oct 2014 23:25:31 +0000 (23:25 +0000)]
[AVX512] Move DAG for all-zero node to X86VectorVTInfo

No functional change.

No change in X86.td.expanded except for the appearance of the new attributes.

The new attributes will be used in the subsequent patch.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219360 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Support mask register in MRMDestReg format
Adam Nemet [Wed, 8 Oct 2014 23:25:29 +0000 (23:25 +0000)]
[AVX512] Support mask register in MRMDestReg format

This is necessary for masking vextract*x4.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219359 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Peel off an asm-only class from AVX512_masking_common.
Adam Nemet [Wed, 8 Oct 2014 23:25:23 +0000 (23:25 +0000)]
[AVX512] Peel off an asm-only class from AVX512_masking_common.

No functional change.

This enables the generation of masking instructions that don't provide a
ISel pattern.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219358 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Don't transform atomic-load-add into an inc/dec when inc/dec is slow
Robin Morisset [Wed, 8 Oct 2014 23:16:23 +0000 (23:16 +0000)]
[X86] Don't transform atomic-load-add into an inc/dec when inc/dec is slow

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219357 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReformat some stuff I missed in recent previous commits
David Blaikie [Wed, 8 Oct 2014 23:09:42 +0000 (23:09 +0000)]
Reformat some stuff I missed in recent previous commits

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219356 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r219354. It seems to break some buildbots.
Alexey Samsonov [Wed, 8 Oct 2014 23:07:59 +0000 (23:07 +0000)]
Revert r219354. It seems to break some buildbots.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219355 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse llvm-symbolizer to symbolize LLVM/Clang crash dumps.
Alexey Samsonov [Wed, 8 Oct 2014 22:57:47 +0000 (22:57 +0000)]
Use llvm-symbolizer to symbolize LLVM/Clang crash dumps.

This change modifies fatal signal handler used in LLVM tools.
Now it attempts to find llvm-symbolizer binary and communicates
with it in order to turn instruction addresses into
function/file/line info entries. This should significantly improve
stack traces readability in Debug builds.

This feature only works on selected platforms (including Darwin
and Linux). If the symbolization fails for some reason, signal
handler will fallback to the original behavior.

Reviewed in http://reviews.llvm.org/D5610

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219354 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSink and coalesce DwarfDebug.cpp::addSectionLabel and DwarfUnit::addSectionLabel...
David Blaikie [Wed, 8 Oct 2014 22:46:27 +0000 (22:46 +0000)]
Sink and coalesce DwarfDebug.cpp::addSectionLabel and DwarfUnit::addSectionLabel down into DwarfCompileUnit::addSectionLabel

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219351 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove dead call to getTypeToTransformTo. The result is
Eric Christopher [Wed, 8 Oct 2014 22:25:45 +0000 (22:25 +0000)]
Remove dead call to getTypeToTransformTo. The result is
unused.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219347 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDebugInfo: The rest of pushing DwarfDebug::constructScopeDIE down into DwarfCompileUnit
David Blaikie [Wed, 8 Oct 2014 22:23:10 +0000 (22:23 +0000)]
DebugInfo: The rest of pushing DwarfDebug::constructScopeDIE down into DwarfCompileUnit

Funnily enough, I copied it, but didn't actually remove the original in
r219345. Let's do that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219346 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPush DwarfDebug::constructScopeDIE down into DwarfCompileUnit
David Blaikie [Wed, 8 Oct 2014 22:20:02 +0000 (22:20 +0000)]
Push DwarfDebug::constructScopeDIE down into DwarfCompileUnit

One of many steps to generalize subprogram emission to both the DWO and
non-DWO sections (to emit -gmlt-like data under fission). Once the
functions are pushed down into DwarfCompileUnit some of the data
structures will be pushed at least into DwarfFile so that they can be
unique per-file, allowing emission to both files independently.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219345 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove a bunch of getSubtargetImpl calls since we already have
Eric Christopher [Wed, 8 Oct 2014 21:08:32 +0000 (21:08 +0000)]
Remove a bunch of getSubtargetImpl calls since we already have
a cached TLI instance.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219342 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Avoid generating inc/dec when slow for x.atomic_store(1 + x.atomic_load())
Robin Morisset [Wed, 8 Oct 2014 19:38:18 +0000 (19:38 +0000)]
[X86] Avoid generating inc/dec when slow for x.atomic_store(1 + x.atomic_load())

Summary:
I had forgotten to check for NotSlowIncDec in the patterns that can generate
inc/dec for the above pattern (added in D4796).
This currently applies to Atom Silvermont, KNL and SKX.

Test Plan: New checks on atomic_mi.ll

Reviewers: jfb, nadav

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5677

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219336 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoInliner: Non-local functions in COMDATs shouldn't be dropped
David Majnemer [Wed, 8 Oct 2014 19:32:32 +0000 (19:32 +0000)]
Inliner: Non-local functions in COMDATs shouldn't be dropped

A function with discardable linkage cannot be discarded if its a member
of a COMDAT group without considering all the other COMDAT members as
well.  This sort of thing is already handled by GlobalOpt/GlobalDCE.

This fixes PR21206.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219335 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix COFF section index relocation should be 16 bits, not 32
Timur Iskhodzhanov [Wed, 8 Oct 2014 18:01:49 +0000 (18:01 +0000)]
Fix COFF section index relocation should be 16 bits, not 32

Original patch by Andrey Guskov!
http://reviews.llvm.org/D5651

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219327 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCorrectly compute the size of common symbols in COFF.
Rafael Espindola [Wed, 8 Oct 2014 17:37:19 +0000 (17:37 +0000)]
Correctly compute the size of common symbols in COFF.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219324 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoPrint symbol sizes in this test in preparation for fixing COFF common sizes.
Rafael Espindola [Wed, 8 Oct 2014 17:19:42 +0000 (17:19 +0000)]
Print symbol sizes in this test in preparation for fixing COFF common sizes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219320 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert "[InstCombine] re-commit r218721 with fix for pr21199"
Justin Bogner [Wed, 8 Oct 2014 16:30:22 +0000 (16:30 +0000)]
Revert "[InstCombine] re-commit r218721 with fix for pr21199"

This seems to cause a miscompile when building clang, which causes a
bootstrapped clang to fail or crash in several of its tests.

See:
  http://lab.llvm.org:8013/builders/clang-x86_64-darwin11-RA/builds/1184
  http://bb.pgr.jp/builders/clang-3stage-x86_64-linux/builds/7813

This reverts commit r219282.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219317 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Added intrinsics for 128-, 256- and 512-bit versions of VPCMP/VPCMPU{BWDQ}
Robert Khasanov [Wed, 8 Oct 2014 15:49:26 +0000 (15:49 +0000)]
[AVX512] Added intrinsics for 128-, 256- and 512-bit versions of VPCMP/VPCMPU{BWDQ}
Added CMP_MASK_CC intrinsic type.
Added tests for intrinsics.

Patch by Sergey Lisitsyn <sergey.lisitsyn@intel.com>

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219316 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove bogus std::error_code returns form SectionRef.
Rafael Espindola [Wed, 8 Oct 2014 15:28:58 +0000 (15:28 +0000)]
Remove bogus std::error_code returns form SectionRef.

There are two methods in SectionRef that can fail:

* getName: The index into the string table can be invalid.
* getContents: The section might point to invalid contents.

Every other method will always succeed and returning and std::error_code just
complicates the code. For example, a section can have an invalid alignment,
but if we are able to get to the section structure at all and create a
SectionRef, we will always be able to read that invalid alignment.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219314 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix indentation.
Rafael Espindola [Wed, 8 Oct 2014 15:12:20 +0000 (15:12 +0000)]
Fix indentation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219312 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate dwarf::ApplePropertyAttributes enum to meaningful values.
Frederic Riss [Wed, 8 Oct 2014 14:59:44 +0000 (14:59 +0000)]
Update dwarf::ApplePropertyAttributes enum to meaningful values.

Summary:
We currently emit an DW_AT_APPLE_property_attribute with a value that is a
bitfield describing the various attributes applied to an ObjectiveC property.
While trying to add testing to one of my dwarfdump patches that would pretty
print that, I realized this information looks totally broken and has maybe
never been correct.

As with every DWARF info, we have some enum in Dwarf.h that describes this
attribute (enum ApplePropertyAttributes). It seems however that the attribute
value is set from another definition of these flags in Sema/DeclSpec.h (enum
ObjCPropertyAttributeKind). And these 2 enums aren't in sync.

This patch updates the Dwarf.h values to the ones we are (and have been for
a very long time) emitting. We change some publicly (and even documented
in SourceLevelDebugging.rst) values, but I doubt this could be an issue as
the information has been wrong for so long...

Reviewers: echristo, dblaikie, aprantl

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5653

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219311 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AVX512] Refactoring of avx512_binop_rm multiclass through AVX512_masking.
Robert Khasanov [Wed, 8 Oct 2014 14:37:45 +0000 (14:37 +0000)]
[AVX512] Refactoring of avx512_binop_rm multiclass through AVX512_masking.
Added new argrument for AVX512_masking: InstrItinClass and bit isCommutable.
No functional change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219310 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove an out of date comment.
Rafael Espindola [Wed, 8 Oct 2014 14:27:05 +0000 (14:27 +0000)]
Remove an out of date comment.

After 4 years there is still no normalization library. We do support
disassembly and relocations, so it doesn't look like we need it.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219308 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoEmit unaligned access build attribute for ARM
Renato Golin [Wed, 8 Oct 2014 12:26:22 +0000 (12:26 +0000)]
Emit unaligned access build attribute for ARM

Patch by Charlie Turner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219301 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRefactor isThumb1Only() && isMClass() into a predicate called isV6M()
Renato Golin [Wed, 8 Oct 2014 12:26:16 +0000 (12:26 +0000)]
Refactor isThumb1Only() && isMClass() into a predicate called isV6M()

This must be enforced for all v6M cores, not just the cortex-m0,
irregardless of the user-specified alignment.

Patch by Charlie Turner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219300 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSimplify switch statement in ARM subtarget align access
Renato Golin [Wed, 8 Oct 2014 12:26:13 +0000 (12:26 +0000)]
Simplify switch statement in ARM subtarget align access

This switch can be reduced to a simpler if/else statement.

Patch by Charlie Turner.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219299 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse the TargetLowering information we already have on the
Eric Christopher [Wed, 8 Oct 2014 09:50:54 +0000 (09:50 +0000)]
Use the TargetLowering information we already have on the
SelectionDAG in SelectionDAGBuilder rather than going through
the TargetMachine for lookup.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219292 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGrab the TargetRegisterInfo off of the subtarget from the
Eric Christopher [Wed, 8 Oct 2014 09:50:52 +0000 (09:50 +0000)]
Grab the TargetRegisterInfo off of the subtarget from the
MachineFunction rather than a lookup on the TargetMachine
to avoid unnecessary lookups.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219291 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUpdate git-svnrevert to accept git and svn revisions
Renato Golin [Wed, 8 Oct 2014 09:32:47 +0000 (09:32 +0000)]
Update git-svnrevert to accept git and svn revisions

Interchangeable commit ids can now be used on this git-svnrevert, which
will figure out what kind of commit that is (if you use format rNNNN for SVN
commits) and make sure the right ids are used in the right places.

It's a little bit more robust and user-friendly.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219290 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUnix/Process: Don't use pthread_sigmask if we aren't built with threads
David Majnemer [Wed, 8 Oct 2014 08:48:43 +0000 (08:48 +0000)]
Unix/Process: Don't use pthread_sigmask if we aren't built with threads

We won't link in pthreads if we weren't built with LLVM_ENABLE_THREADS
which means we won't get access to pthread_sigmask.  Use sigprocmask
instead.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219288 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReplace calls to get the subtarget and TargetFrameLowering with
Eric Christopher [Wed, 8 Oct 2014 08:46:34 +0000 (08:46 +0000)]
Replace calls to get the subtarget and TargetFrameLowering with
cached variables and a single call in the constructor.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219287 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFormat spacing and remove extra lines to comply with standards. NFC.
Suyog Sarda [Wed, 8 Oct 2014 08:37:49 +0000 (08:37 +0000)]
Format spacing and remove extra lines to comply with standards. NFC.

Differential Revision: http://reviews.llvm.org/D5649

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219286 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoUse cached subtarget rather than looking it up on the
Eric Christopher [Wed, 8 Oct 2014 07:51:41 +0000 (07:51 +0000)]
Use cached subtarget rather than looking it up on the
TargetMachine again.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219285 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCache TargetLowering on SelectionDAGISel and update previous
Eric Christopher [Wed, 8 Oct 2014 07:32:17 +0000 (07:32 +0000)]
Cache TargetLowering on SelectionDAGISel and update previous
calls to getTargetLowering() with the cached variable.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219284 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGlobalOpt: Don't drop unused memberes of a Comdat
David Majnemer [Wed, 8 Oct 2014 07:23:31 +0000 (07:23 +0000)]
GlobalOpt: Don't drop unused memberes of a Comdat

A linkonce_odr member of a COMDAT shouldn't be dropped if we need to
keep the entire COMDAT group.

This fixes PR21191.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219283 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine] re-commit r218721 with fix for pr21199
Gerolf Hoflehner [Wed, 8 Oct 2014 06:42:19 +0000 (06:42 +0000)]
[InstCombine] re-commit r218721 with fix for pr21199

The icmp-select-icmp optimization targets select-icmp.eq
only. This is now ensured by testing the branch predicate
explictly. This commit also includes the test case for pr21199.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219282 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCOFF: Don't oversize COMMON symbols when targeting BFD ld
David Majnemer [Wed, 8 Oct 2014 06:38:53 +0000 (06:38 +0000)]
COFF: Don't oversize COMMON symbols when targeting BFD ld

COFF normally doesn't allow us to describe the alignment of COMMON
symbols.

It turns out that most linkers use the symbol size as a hint as to how
aligned the symbol should be.

However the BFD folks have added a .drectve command, which we
now support as of r219229, that allows us to specify the alignment
precisely.  With this in mind, stop rounding sizes up.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219281 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm-dwarfdump: Add support for some COFF relocations
David Majnemer [Wed, 8 Oct 2014 06:38:50 +0000 (06:38 +0000)]
llvm-dwarfdump: Add support for some COFF relocations

DWARF in COFF utilizes several relocations.  Implement support for them
in RelocVisitor to support llvm-dwarfdump.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219280 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[AArch64] Generate vector signed/unsigned mul and mla/mls long.
Chad Rosier [Wed, 8 Oct 2014 02:31:24 +0000 (02:31 +0000)]
[AArch64] Generate vector signed/unsigned mul and mla/mls long.

Phabricator Revision: http://reviews.llvm.org/D5589
Patch by Balaram Makam <bmakam@codeaurora.org>!!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219276 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm-readobj: add test for r219228
Rui Ueyama [Wed, 8 Oct 2014 02:06:11 +0000 (02:06 +0000)]
llvm-readobj: add test for r219228

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219274 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoCache SelectionDAGISel TargetInstrInfo lookups on the class and
Eric Christopher [Wed, 8 Oct 2014 01:58:03 +0000 (01:58 +0000)]
Cache SelectionDAGISel TargetInstrInfo lookups on the class and
propagate. Also use the TargetSubtargetInfo and the MachineFunction
and move TargetRegisterInfo query closer to uses.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219273 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReset the target options and optimization level as the first
Eric Christopher [Wed, 8 Oct 2014 01:58:01 +0000 (01:58 +0000)]
Reset the target options and optimization level as the first
thing we do inside selection dag. This code needs to be
migrated to queries on the function rather than global
data, but this organizes things before we start grabbing
the subtarget.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219271 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHave the selection dag grab TargetLowering off of the subtarget
Eric Christopher [Wed, 8 Oct 2014 01:57:58 +0000 (01:57 +0000)]
Have the selection dag grab TargetLowering off of the subtarget
inside init rather than have it passed in as an argument.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219270 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRevert r219175 - [InstCombine] re-commit r218721 icmp-select-icmp optimization
Hans Wennborg [Wed, 8 Oct 2014 01:05:57 +0000 (01:05 +0000)]
Revert r219175 - [InstCombine] re-commit r218721 icmp-select-icmp optimization

This seems to have caused PR21199.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219264 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoHave SelectionDAG's subtarget TargetSelectionDAGInfo be set
Eric Christopher [Wed, 8 Oct 2014 00:32:59 +0000 (00:32 +0000)]
Have SelectionDAG's subtarget TargetSelectionDAGInfo be set
during init rather than construction time.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219262 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DebugInfo] Pass DWARFSection down to DWARFUnit constructor (NFC).
Alexey Samsonov [Wed, 8 Oct 2014 00:24:41 +0000 (00:24 +0000)]
[DebugInfo] Pass DWARFSection down to DWARFUnit constructor (NFC).

Keep the actual section contents and the relocation map together.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219261 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Support] Add MemoryBuffer::getFileSlice()
Nick Kledzik [Wed, 8 Oct 2014 00:22:18 +0000 (00:22 +0000)]
[Support] Add MemoryBuffer::getFileSlice()

mach-o supports "fat" files which are a header/table-of-contents followed by a
concatenation of mach-o files built for different architectures. Currently,
MemoryBuffer has no easy way to map a subrange (slice) of a file which lld
will need to select a mach-o slice of a fat file. The new function provides
an easy way to map a slice of a file into a MemoryBuffer. Test case included.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219260 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DebugInfo] Pass DWARFSection into DWARFUnitSection constructor (NFC).
Alexey Samsonov [Wed, 8 Oct 2014 00:07:53 +0000 (00:07 +0000)]
[DebugInfo] Pass DWARFSection into DWARFUnitSection constructor (NFC).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219259 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Fix a bug with fetch_add(INT32_MIN)
Robin Morisset [Tue, 7 Oct 2014 23:53:57 +0000 (23:53 +0000)]
[X86] Fix a bug with fetch_add(INT32_MIN)

Summary:
Fix pr21099

The pseudocode of what we were doing (spread through two functions) was:
if (operand.doesNotFitIn32Bits())
  Opc.initializeWithFoo();
if (operand < 0)
  operand = -operand;
if (operand.doesFitIn8Bits())
  Opc.initializeWithBar();
else if (operand.doesFitIn32Bits())
  Opc.initializeWithBlah();
doStuff(Opc);

So for operand == INT32_MIN, Opc was never initialized because the operand changes
from fitting in 32 bits to not fitting, causing the various bugs/error messages
noted by pr21099.

This patch adds an extra test at the beginning for this case, and an
llvm_unreachable to have better error message if the operand ends up
not fitting in 32-bits at the end.

Test Plan: new test + make check

Reviewers: jfb

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5655

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219257 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Refactor VOP3 instruction defs
Tom Stellard [Tue, 7 Oct 2014 23:51:41 +0000 (23:51 +0000)]
R600/SI: Refactor VOP3 instruction defs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219256 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Refactor VOPC instruction defs
Tom Stellard [Tue, 7 Oct 2014 23:51:39 +0000 (23:51 +0000)]
R600/SI: Refactor VOPC instruction defs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219255 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Refactor VOP2 instruction defs
Tom Stellard [Tue, 7 Oct 2014 23:51:38 +0000 (23:51 +0000)]
R600/SI: Refactor VOP2 instruction defs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219254 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Refactor VOP1 instruction defs
Tom Stellard [Tue, 7 Oct 2014 23:51:34 +0000 (23:51 +0000)]
R600/SI: Refactor VOP1 instruction defs

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219253 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[DebugInfo] Turn DWARFContext::Section into DWARFSection (NFC).
Alexey Samsonov [Tue, 7 Oct 2014 23:45:11 +0000 (23:45 +0000)]
[DebugInfo] Turn DWARFContext::Section into DWARFSection (NFC).

It would be more convenient to pass DWARFSection into DWARFUnitSection
constructor, instead of passing its components (Data and RelocAddrMap)
as a separate arguments.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219252 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd return value and negative checks to MapVector::erase from r219240.
Kaelyn Takata [Tue, 7 Oct 2014 23:11:49 +0000 (23:11 +0000)]
Add return value and negative checks to MapVector::erase from r219240.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219250 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDebugInfo+DFSan: Ensure that debug info references to llvm::Functions remain pointing...
David Blaikie [Tue, 7 Oct 2014 22:59:46 +0000 (22:59 +0000)]
DebugInfo+DFSan: Ensure that debug info references to llvm::Functions remain pointing to the underlying function when wrappers are created

This is somewhat the inverse of how similar bugs in DAE and ArgPromo
manifested and were addressed. In those passes, individual call sites
were visited explicitly, and then the old function was deleted. This
left the debug info with a null llvm::Function* that needed to be
updated to point to the new function.

In the case of DFSan, it RAUWs the old function with the wrapper, which
includes debug info. So now the debug info refers to the wrapper, which
doesn't actually have any instructions with debug info in it, so it is
ignored entirely - resulting in a DW_TAG_subprogram with no high/low pc,
etc. Instead, fix up the debug info to refer to the original function
after the RAUW messed it up.

Reviewed/discussed with Peter Collingbourne on the llvm-dev mailing
list.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219249 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[Bugpoint] Close error log in ProcessFailure. Bugpoint had been failing to close
Lang Hames [Tue, 7 Oct 2014 21:47:23 +0000 (21:47 +0000)]
[Bugpoint] Close error log in ProcessFailure. Bugpoint had been failing to close
this, and in some circumstances (e.g. reducing particularly large test-cases)
this was causing bugpoint to be killed for hitting open file-handle limits.

No test case: I was only able to trigger this with test cases taking upwards of
10 mins to run.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219244 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Remove dead code
Matt Arsenault [Tue, 7 Oct 2014 21:29:56 +0000 (21:29 +0000)]
R600: Remove dead code

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219242 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLoopUnroll: Create sub-loops in LoopInfo
Duncan P. N. Exon Smith [Tue, 7 Oct 2014 21:19:00 +0000 (21:19 +0000)]
LoopUnroll: Create sub-loops in LoopInfo

`LoopUnrollPass` says that it preserves `LoopInfo` -- make it so.  In
particular, tell `LoopInfo` about copies of inner loops when unrolling
the outer loop.

Conservatively, also tell `ScalarEvolution` to forget about the original
versions of these loops, since their inputs may have changed.

Fixes PR20987.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219241 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAdd size_t MapVector::erase(KeyT) similar to the one in std::map.
Kaelyn Takata [Tue, 7 Oct 2014 21:15:51 +0000 (21:15 +0000)]
Add size_t MapVector::erase(KeyT) similar to the one in std::map.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219240 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoLoopUnroll: Only check for ScalarEvolution analysis once, NFC
Duncan P. N. Exon Smith [Tue, 7 Oct 2014 21:12:44 +0000 (21:12 +0000)]
LoopUnroll: Only check for ScalarEvolution analysis once, NFC

A follow-up commit will add use to a tight loop.  We might as well just
find it once anyway.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219239 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Remove some redundant initializations from AMDGPUMCAsmInfo
Tom Stellard [Tue, 7 Oct 2014 21:09:25 +0000 (21:09 +0000)]
R600: Remove some redundant initializations from AMDGPUMCAsmInfo

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219238 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600: Use MCAsmInfoELF as AMDGPUMCAsmInfo base class
Tom Stellard [Tue, 7 Oct 2014 21:09:23 +0000 (21:09 +0000)]
R600: Use MCAsmInfoELF as AMDGPUMCAsmInfo base class

The main reason for this is that the MCAsmInfo class,
which we were previously using as the base class, sets
PrivateGlobalPrefix to "L", which causes all global
functions that start with L to be treated as local symbols.

MCAsmInfoELF sets PrivateGlobalPrefix to ".L", which is what
we want, and it is probably a good idea to use this as the
base class anyway, since we are emitting ELF binaries.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219237 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoR600/SI: Remove assertion in SIInstrInfo::areLoadsFromSameBasePtr()
Tom Stellard [Tue, 7 Oct 2014 21:09:20 +0000 (21:09 +0000)]
R600/SI: Remove assertion in SIInstrInfo::areLoadsFromSameBasePtr()

Added a FIXME coment instead, we need to handle the case where the
two DS instructions being compared have different numbers of operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219236 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDon't check for null after calling COFFObjectFile::toSec.
Rafael Espindola [Tue, 7 Oct 2014 21:03:22 +0000 (21:03 +0000)]
Don't check for null after calling COFFObjectFile::toSec.

It can only return null if passed a corrupted reference with a null Ref.p.
Checking for null is then an issue for asserts to check for internal
consistency, not control flow to check for invalid input.

I didn't add an assert(sec != nullptr) because toSec itself has a far more
complete assert.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219235 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoOptimize COFFObjectFile::sectionContainsSymbol a bit.
Rafael Espindola [Tue, 7 Oct 2014 20:42:47 +0000 (20:42 +0000)]
Optimize COFFObjectFile::sectionContainsSymbol a bit.

There is no need to compute the coff_section of the symbol just to compare the
pointer.

Inspired by the ELF implementation.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219233 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoMC: add support for -aligncomm GNU extension
Saleem Abdulrasool [Tue, 7 Oct 2014 19:37:57 +0000 (19:37 +0000)]
MC: add support for -aligncomm GNU extension

The GNU linker supports an -aligncomm directive that allows for power-of-2
alignment of common data.  Add support to emit this directive.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219229 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agollvm-readobj: add support to dump (COFF) directives
Saleem Abdulrasool [Tue, 7 Oct 2014 19:37:52 +0000 (19:37 +0000)]
llvm-readobj: add support to dump (COFF) directives

PE/COFF has a special section (.drectve) which can be used to pass options to
the linker (similar to LC_LINKER_OPTION).  Add support to llvm-readobj to print
the contents of the section for tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219228 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove the IsVolatileSize parameter of getOpenFileSlice.
Rafael Espindola [Tue, 7 Oct 2014 19:09:05 +0000 (19:09 +0000)]
Remove the IsVolatileSize parameter of getOpenFileSlice.

getOpenFileSlice gets passed the map size, so it makes no sense to say that
the size is volatile. The code will not even compute the size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219226 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoBe consistent about using "const Twine &" for filenames.
Rafael Espindola [Tue, 7 Oct 2014 18:58:55 +0000 (18:58 +0000)]
Be consistent about using "const Twine &" for filenames.

On this file we had a mix of
* Twine
* const char *
* StringRef

The two that make sense are
* const Twine & (caller convenience)
* consc char * (that is what will eventually be passed to open.

Given that sys::fs::openFileForRead takes a "const Twine &", I picked that.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219224 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoTwo case switch to select optimization
Marcello Maggioni [Tue, 7 Oct 2014 18:16:44 +0000 (18:16 +0000)]
Two case switch to select optimization

This optimization tries to convert switch instructions that are used to select a value with only 2 unique cases + default block
to a select or a couple of selects (depending if the default block is reachable or not).

The typical case this optimization wants to be able to optimize is this one:

Example:
switch (a) {
  case 10:                %0 = icmp eq i32 %a, 10
    return 10;            %1 = select i1 %0, i32 10, i32 4
  case 20:        ---->   %2 = icmp eq i32 %a, 20
    return 2;             %3 = select i1 %2, i32 2, i32 %1
  default:
    return 4;
}

It also sets the base for further optimizations that are planned and being reviewed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219223 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agotypos
Sanjay Patel [Tue, 7 Oct 2014 17:38:33 +0000 (17:38 +0000)]
typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219221 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agotypos
Sanjay Patel [Tue, 7 Oct 2014 17:36:50 +0000 (17:36 +0000)]
typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219220 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agodon't repeat function name in comments
Sanjay Patel [Tue, 7 Oct 2014 17:28:04 +0000 (17:28 +0000)]
don't repeat function name in comments

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219218 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agotypos
Sanjay Patel [Tue, 7 Oct 2014 17:00:16 +0000 (17:00 +0000)]
typos

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219216 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDebugInfo: Ensure that all debug location scope chains from instructions within a...
David Blaikie [Tue, 7 Oct 2014 16:56:20 +0000 (16:56 +0000)]
DebugInfo: Ensure that all debug location scope chains from instructions within a function, lead to the function itself.

Let me tell you a tale...

Originally committed in r211723 after discovering a nasty case of weird
scoping due to inlining, this was reverted in r211724 after it fired in
ASan/compiler-rt.

(minor diversion where I accidentally committed/reverted again in
r211871/r211873)

After further testing and fixing bugs in ArgumentPromotion (r211872) and
Inlining (r212065) it was recommitted in r212085. Reverted in r212089
after the sanitizer buildbots still showed problems.

Fixed another bug in ArgumentPromotion (r212128) found by this
assertion.

Recommitted in r212205, reverted in r212226 after it crashed some more
on sanitizer buildbots.

Fix clang some more in r212761.

Recommitted in r212776, reverted in r212793. ASan failures.
Recommitted in r213391, reverted in r213432, trying to reproduce flakey
ASan build failure.

Fixed bugs in r213805 (ArgPromo + DebugInfo), r213952
(LiveDebugVariables strips dbg_value intrinsics in functions not
described by debug info).

Recommitted in r214761, reverted in r214999, flakey failure on Windows
buildbot.

Fixed DeadArgElimination + DebugInfo bug in r219210.

Recommitting and hoping that's the last of it.

[That one burned down, fell over, then sank into the swamp.]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219215 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoDebugInfo+DeadArgElimination: Ensure llvm::Function*s from debug info are updated...
David Blaikie [Tue, 7 Oct 2014 15:10:23 +0000 (15:10 +0000)]
DebugInfo+DeadArgElimination: Ensure llvm::Function*s from debug info are updated even when DAE removes both varargs and non-varargs arguments on the same function.

After some stellar (& inspired) help from Reid Kleckner providing a test
case for some rather unstable undefined behavior showing up as
assertions produced by r214761, I was able to fix this issue in DAE
involving the application of both varargs removal, followed by normal
argument removal.

Indeed I introduced this same bug into ArgumentPromotion (r212128) by
copying the code from DAE, and when I fixed the bug in ArgPromo
(r213805) and commented in that patch that I didn't need to address the
same issue in DAE because it was a single pass. Turns out it's two pass,
one for the varargs and one for the normal arguments, so the same fix is
needed (at least during varargs removal). So here it is.

(the observable/net effect of this bug, even when it didn't result in
assertion failure, is that debug info would describe the DAE'd function
in the abstract, but wouldn't provide high/low_pc, variable locations,
line table, etc (it would appear as though the function had been
entirely optimized away), see the original PR14016 for details of the
general problem)

I'm not recommitting the assertion just yet, as there's been another
regression of it since I last tried. It might just be a few test cases
weren't adequately updated after Adrian or Duncan's recent schema
changes.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219210 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReformat if statement to comply with LLVM standards. NFC.
Suyog Sarda [Tue, 7 Oct 2014 12:04:07 +0000 (12:04 +0000)]
Reformat if statement to comply with LLVM standards. NFC.

Differential Revision: http://reviews.llvm.org/D5644

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219203 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoReformat to comply with LLVM coding standards using clang-format.
Suyog Sarda [Tue, 7 Oct 2014 11:56:06 +0000 (11:56 +0000)]
Reformat to comply with LLVM coding standards using clang-format.
NFC.

Differential Revision: http://reviews.llvm.org/D5645

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219202 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoRemove Extra lines. NFC.
Suyog Sarda [Tue, 7 Oct 2014 11:31:31 +0000 (11:31 +0000)]
Remove Extra lines. NFC.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219201 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[asan-asm-instrumentation] CFI directives are generated for .S files.
Yuri Gorshenin [Tue, 7 Oct 2014 11:03:09 +0000 (11:03 +0000)]
[asan-asm-instrumentation] CFI directives are generated for .S files.

Summary: CFI directives are generated for .S files.

Reviewers: eugenis

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5520

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219199 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[InstCombine] Reformat if statements to comply with LLVM Coding Standards.
Tilmann Scheller [Tue, 7 Oct 2014 10:19:34 +0000 (10:19 +0000)]
[InstCombine] Reformat if statements to comply with LLVM Coding Standards.

Patch by Sonam Kumari!

Differential Revision: http://reviews.llvm.org/D5643

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219198 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[mips] Return {f128} correctly for N32/N64.
Daniel Sanders [Tue, 7 Oct 2014 09:29:59 +0000 (09:29 +0000)]
[mips] Return {f128} correctly for N32/N64.

Summary:
According to the ABI documentation, f128 and {f128} should both be returned
in $f0 and $f2. However, this doesn't match GCC's behaviour which is to
return f128 in $f0 and $f2, but {f128} in $f0 and $f1.

Reviewers: vmedic

Reviewed By: vmedic

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D5578

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219196 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[X86] Fix a bug where the disassembler was ignoring the VEX.W bit in 32-bit mode...
Craig Topper [Tue, 7 Oct 2014 07:29:50 +0000 (07:29 +0000)]
[X86] Fix a bug where the disassembler was ignoring the VEX.W bit in 32-bit mode for certain instructions it shouldn't.

Unfortunately, this isn't easy to fix since there's no simple way to figure out from the disassembler tables whether the W-bit is being used to select a 64-bit GPR or if its a required part of the opcode. The fix implemented here just looks for "64" in the instruction name and ignores the W-bit in 32-bit mode if its present.

Fixes PR21169.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219194 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFormatting fixes. Most putting 'else' on the same line as the preceding curly brace.
Craig Topper [Tue, 7 Oct 2014 07:29:48 +0000 (07:29 +0000)]
Formatting fixes. Most putting 'else' on the same line as the preceding curly brace.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219193 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoFix filename in header and use C++ version of the C header files.
Craig Topper [Tue, 7 Oct 2014 07:29:46 +0000 (07:29 +0000)]
Fix filename in header and use C++ version of the C header files.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219192 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoGlobalDCE: Don't drop any COMDAT members
David Majnemer [Tue, 7 Oct 2014 07:07:19 +0000 (07:07 +0000)]
GlobalDCE: Don't drop any COMDAT members

If we require a single member of a comdat, require all of the other
members as well.

This fixes PR20981.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219191 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoAttempt to calm down buildbots
David Majnemer [Tue, 7 Oct 2014 05:56:45 +0000 (05:56 +0000)]
Attempt to calm down buildbots

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219190 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agoSupport: Don't call close again if we get EINTR
David Majnemer [Tue, 7 Oct 2014 05:48:40 +0000 (05:48 +0000)]
Support: Don't call close again if we get EINTR

Most Unix-like operating systems guarantee that the file descriptor is
closed after a call to close(2), even if close comes back with EINTR.
For these systems, calling close _again_ will either do nothing or close
some other file descriptor open(2)'d by another thread. (Linux)

However, some operating systems do not have this behavior.  They require
at least another call to close(2) before guaranteeing that the
descriptor is closed. (HP-UX)

And some operating systems have an unpredictable blend of the two
behaviors! (xnu)

Avoid this disaster by blocking all signals before we call close(2).
This ensures that a signal will not be delivered to the thread and
close(2) will not give us back EINTR.  We restore the signal mask once
the operation is done.

N.B. This isn't a problem on Windows, it doesn't have a notion of EINTR
because signals always get delivered to dedicated signal handling
threads.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219189 91177308-0d34-0410-b5e6-96231b3b80d8

9 years agogold plugin: Handle gold selecting a linkonce GV when a weak is present.
Rafael Espindola [Tue, 7 Oct 2014 04:06:13 +0000 (04:06 +0000)]
gold plugin: Handle gold selecting a linkonce GV when a weak is present.

The plugin API doesn't have the notion of linkonce, only weak. It is up to the
plugin to figure out if a symbol used only for the symbol table can be dropped.
In particular, it has to avoid dropping a linkonce_odr selected by gold if there
is also a weak_odr.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219188 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][AArch64] Teach the address computation code to also fold sign-/zero-extends.
Juergen Ributzka [Tue, 7 Oct 2014 03:40:06 +0000 (03:40 +0000)]
[FastISel][AArch64] Teach the address computation code to also fold sign-/zero-extends.

The code already folds sign-/zero-extends, but only if they are arguments to
mul and shift instructions. This extends the code to also fold them when they
are direct inputs.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219187 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][AArch64] Teach the address computation to also fold sub instructions.
Juergen Ributzka [Tue, 7 Oct 2014 03:40:03 +0000 (03:40 +0000)]
[FastISel][AArch64] Teach the address computation to also fold sub instructions.

Tiny enhancement to the address computation code to also fold sub instructions
if the rhs is constant and can be folded into the offset.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219186 91177308-0d34-0410-b5e6-96231b3b80d8

9 years ago[FastISel][AArch64] Fix "Fold sign-/zero-extends into the load instruction."
Juergen Ributzka [Tue, 7 Oct 2014 03:39:59 +0000 (03:39 +0000)]
[FastISel][AArch64] Fix "Fold sign-/zero-extends into the load instruction."

This commit fixes an issue with sign-/zero-extending loads that was discovered
by Richard Barton.

We use now the correct load instructions for sign-extending loads to 64bit. Also
updated and added more unit tests.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@219185 91177308-0d34-0410-b5e6-96231b3b80d8